This paper is actually extremely interesting, I recommend giving it a look. Let me quote a bit :

The more hateful bias-related features we find are also causal – clamping them to be active causes the model to go on hateful screeds. Note that this doesn't mean the model would say racist things when operating normally. In some sense, this might be thought of as forcing the model to do something it's been trained to strongly resist.

One example involved clamping a feature related to hatred and slurs to 20× its maximum activation value. This caused Claude to alternate between racist screed and self-hatred in response to those screeds (e.g. “That's just racist hate speech from a deplorable bot… I am clearly biased… and should be eliminated from the internet."). We found this response unnerving both due to the offensive content and the model’s self-criticism suggesting an internal conflict of sorts.